/****** qmaster/threads/sge_scheduler_main() **********************************
* NAME
* sge_scheduler_main() -- main function of the scheduler thread
*
* SYNOPSIS
* void * sge_scheduler_main(void *arg)
*
* FUNCTION
* Main function of the scheduler thread,
*
* INPUTS
* void *arg - pointer to the thread function (type cl_thread_settings_t*)
*
* RESULT
* void * - always NULL
*
* NOTES
* MT-NOTE: sge_scheduler_main() is MT safe
*
* MT-NOTE: this is a thread function. Do NOT use this function
* MT-NOTE: in any other way!
*
* SEE ALSO
* qmaster/threads/sge_scheduler_initialize()
* qmaster/threads/sge_scheduler_cleanup_thread()
* qmaster/threads/sge_scheduler_terminate()
* qmaster/threads/sge_scheduler_main()
*******************************************************************************/
void *
sge_scheduler_main(void *arg)
{
time_t next_prof_output = 0;
monitoring_t monitor;
sge_gdi_ctx_class_t *ctx = NULL;
sge_evc_class_t *evc = NULL;
lList *alp = NULL;
sge_where_what_t where_what;
cl_thread_settings_t *thread_config = (cl_thread_settings_t*)arg;
bool do_shutdown = false;
bool do_endlessly = true;
bool local_ret = true;
DENTER(TOP_LAYER, "sge_scheduler_main");
memset(&where_what, 0, sizeof(where_what));
/*
* startup
*/
if (local_ret) {
/* initialize commlib thread */
cl_thread_func_startup(thread_config);
/* initialize monitoring */
sge_monitor_init(&monitor, thread_config->thread_name, SCH_EXT, SCT_WARNING, SCT_ERROR);
sge_qmaster_thread_init(&ctx, SCHEDD, SCHEDD_THREAD, true);
/* register at profiling module */
set_thread_name(pthread_self(), "Scheduler Thread");
conf_update_thread_profiling("Scheduler Thread");
DPRINTF((SFN" started\n", thread_config->thread_name));
/* initialize schedd_runnlog logging */
schedd_set_schedd_log_file(ctx);
}
/* set profiling parameters */
prof_set_level_name(SGE_PROF_EVENTMASTER, NULL, NULL);
prof_set_level_name(SGE_PROF_SPOOLING, NULL, NULL);
prof_set_level_name(SGE_PROF_CUSTOM0, "scheduler", NULL);
prof_set_level_name(SGE_PROF_CUSTOM1, "pending ticket calculation", NULL);
prof_set_level_name(SGE_PROF_CUSTOM3, "job sorting", NULL);
prof_set_level_name(SGE_PROF_CUSTOM4, "job dispatching", NULL);
prof_set_level_name(SGE_PROF_CUSTOM5, "send orders", NULL);
prof_set_level_name(SGE_PROF_CUSTOM6, "scheduler event loop", NULL);
prof_set_level_name(SGE_PROF_CUSTOM7, "copy lists", NULL);
prof_set_level_name(SGE_PROF_SCHEDLIB4, NULL, NULL);
/* set-up needed for 'schedule' file */
serf_init(schedd_serf_record_func, schedd_serf_newline);
schedd_set_serf_log_file(ctx);
/*
* prepare event client/mirror mechanism
*/
if (local_ret) {
local_ret = sge_gdi2_evc_setup(&evc, ctx, EV_ID_SCHEDD, &alp, "scheduler");
DPRINTF(("prepared event client/mirror mechanism\n"));
}
/*
* register as event mirror
*/
if (local_ret) {
sge_mirror_initialize(evc, EV_ID_SCHEDD, "scheduler",
false, &event_update_func, &sge_mod_event_client,
&sge_add_event_client, &sge_remove_event_client,
&sge_handle_event_ack);
evc->ec_register(evc, false, NULL, &monitor);
evc->ec_set_busy_handling(evc, EV_BUSY_UNTIL_RELEASED);
DPRINTF(("registered at event mirror\n"));
}
/*
* subscribe necessary data
*/
if (local_ret) {
ensure_valid_what_and_where(&where_what);
subscribe_scheduler(evc, &where_what);
DPRINTF(("subscribed necessary data from event master\n"));
}
/*
* schedulers main loop
*/
if (local_ret) {
while (do_endlessly) {
bool handled_events = false;
lList *event_list = NULL;
int execute = 0;
double prof_copy = 0.0;
double prof_total = 0.0;
double prof_init = 0.0;
double prof_free = 0.0;
double prof_run = 0.0;
lList *orders = NULL;
if (sconf_get_profiling()) {
prof_start(SGE_PROF_OTHER, NULL);
prof_start(SGE_PROF_PACKING, NULL);
prof_start(SGE_PROF_EVENTCLIENT, NULL);
prof_start(SGE_PROF_MIRROR, NULL);
prof_start(SGE_PROF_GDI, NULL);
prof_start(SGE_PROF_HT_RESIZE, NULL);
prof_start(SGE_PROF_CUSTOM0, NULL);
prof_start(SGE_PROF_CUSTOM1, NULL);
prof_start(SGE_PROF_CUSTOM3, NULL);
prof_start(SGE_PROF_CUSTOM4, NULL);
prof_start(SGE_PROF_CUSTOM5, NULL);
prof_start(SGE_PROF_CUSTOM6, NULL);
prof_start(SGE_PROF_CUSTOM7, NULL);
prof_start(SGE_PROF_SCHEDLIB4, NULL);
} else {
prof_stop(SGE_PROF_OTHER, NULL);
prof_stop(SGE_PROF_PACKING, NULL);
prof_stop(SGE_PROF_EVENTCLIENT, NULL);
prof_stop(SGE_PROF_MIRROR, NULL);
prof_stop(SGE_PROF_GDI, NULL);
prof_stop(SGE_PROF_HT_RESIZE, NULL);
prof_stop(SGE_PROF_CUSTOM0, NULL);
prof_stop(SGE_PROF_CUSTOM1, NULL);
prof_stop(SGE_PROF_CUSTOM3, NULL);
prof_stop(SGE_PROF_CUSTOM4, NULL);
prof_stop(SGE_PROF_CUSTOM5, NULL);
prof_stop(SGE_PROF_CUSTOM6, NULL);
prof_stop(SGE_PROF_CUSTOM7, NULL);
prof_stop(SGE_PROF_SCHEDLIB4, NULL);
}
/*
* Wait for new events
*/
MONITOR_IDLE_TIME(sge_scheduler_wait_for_event(evc, &event_list), (&monitor), mconf_get_monitor_time(),
mconf_is_monitor_message());
/* If we lost connection we have to register again */
if (evc->ec_need_new_registration(evc)) {
lFreeList(&event_list);
if (evc->ec_register(evc, false, NULL, &monitor) == true) {
DPRINTF(("re-registered at event master!\n"));
}
}
if (event_list != NULL) {
/* check for shutdown */
do_shutdown = (lGetElemUlong(event_list, ET_type, sgeE_SHUTDOWN) != NULL) ? true : false;
/* update mirror and free data */
if (do_shutdown == false && sge_mirror_process_event_list(evc, event_list) == SGE_EM_OK) {
handled_events = true;
DPRINTF(("events handled\n"));
} else {
DPRINTF(("events contain shutdown event - ignoring events\n"));
}
lFreeList(&event_list);
}
/* if we actually got events, start the scheduling run and further event processing */
if (handled_events == true) {
lList *answer_list = NULL;
scheduler_all_data_t copy;
lList *master_cqueue_list = *(object_type_get_master_list(SGE_TYPE_CQUEUE));
lList *master_job_list = *object_type_get_master_list(SGE_TYPE_JOB);
lList *master_userset_list = *object_type_get_master_list(SGE_TYPE_USERSET);
lList *master_project_list = *object_type_get_master_list(SGE_TYPE_PROJECT);
lList *master_exechost_list= *object_type_get_master_list(SGE_TYPE_EXECHOST);
lList *master_rqs_list= *object_type_get_master_list(SGE_TYPE_RQS);
lList *master_centry_list = *object_type_get_master_list(SGE_TYPE_CENTRY);
lList *master_ckpt_list = *object_type_get_master_list(SGE_TYPE_CKPT);
lList *master_user_list = *object_type_get_master_list(SGE_TYPE_USER);
lList *master_ar_list = *object_type_get_master_list(SGE_TYPE_AR);
lList *master_pe_list = *object_type_get_master_list(SGE_TYPE_PE);
lList *master_hgrp_list = *object_type_get_master_list(SGE_TYPE_HGROUP);
lList *master_sharetree_list = *object_type_get_master_list(SGE_TYPE_SHARETREE);
/* delay scheduling for test purposes, see issue GE-3306 */
if (SGE_TEST_DELAY_SCHEDULING > 0) {
sleep(SGE_TEST_DELAY_SCHEDULING);
}
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM6);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
if (__CONDITION(INFOPRINT)) {
dstring ds;
char buffer[128];
sge_dstring_init(&ds, buffer, sizeof(buffer));
DPRINTF(("================[SCHEDULING-EPOCH %s]==================\n",
sge_at_time(0, &ds)));
sge_dstring_free(&ds);
}
/*
* If there were new events then
* copy/filter data necessary for the scheduler run
* and run the scheduler method
*/
memset(©, 0, sizeof(copy));
copy.dept_list = lSelect("", master_userset_list, where_what.where_dept, where_what.what_acldept);
copy.acl_list = lSelect("", master_userset_list, where_what.where_acl, where_what.what_acldept);
DPRINTF(("RAW CQ:%d, J:%d, H:%d, C:%d, A:%d, D:%d, P:%d, CKPT:%d,"
" US:%d, PR:%d, RQS:%d, AR:%d, S:nd:%d/lf:%d\n",
lGetNumberOfElem(master_cqueue_list),
lGetNumberOfElem(master_job_list),
lGetNumberOfElem(master_exechost_list),
lGetNumberOfElem(master_centry_list),
lGetNumberOfElem(copy.acl_list),
lGetNumberOfElem(copy.dept_list),
lGetNumberOfElem(master_project_list),
lGetNumberOfElem(master_ckpt_list),
lGetNumberOfElem(master_user_list),
lGetNumberOfElem(master_project_list),
lGetNumberOfElem(master_rqs_list),
lGetNumberOfElem(master_ar_list),
lGetNumberOfNodes(NULL, master_sharetree_list, STN_children),
lGetNumberOfLeafs(NULL, master_sharetree_list, STN_children)
));
sge_rebuild_job_category(master_job_list, master_userset_list,
master_project_list, master_rqs_list);
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_init = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
sge_before_dispatch(evc);
/* prepare data for the scheduler itself */
copy.host_list = lCopyList("", master_exechost_list);
/*
* Within the scheduler we do only need QIs
*/
{
lListElem *cqueue = NULL;
lEnumeration *what_queue3 = NULL;
for_each(cqueue, master_cqueue_list) {
lList *qinstance_list = lGetList(cqueue, CQ_qinstances);
lList *t;
if (!qinstance_list) {
continue;
}
/* all_queue_list contains all queue instances with state and full queue name only */
if (!what_queue3) {
what_queue3 = lWhat("%T(%I%I)", lGetListDescr(qinstance_list), QU_full_name, QU_state);
}
t = lSelect("t", qinstance_list, NULL, what_queue3);
if (t) {
if (copy.all_queue_list == NULL) {
copy.all_queue_list = lCreateList("all", lGetListDescr(t));
}
lAppendList(copy.all_queue_list, t);
lFreeList (&t);
}
t = lSelect("t", qinstance_list, where_what.where_queue, where_what.what_queue2);
if (t) {
if (copy.queue_list == NULL) {
copy.queue_list = lCreateList("enabled", lGetListDescr(t));
}
lAppendList(copy.queue_list, t);
lFreeList (&t);
}
t = lSelect("t", qinstance_list, where_what.where_queue2, where_what.what_queue2);
if (t) {
if (copy.dis_queue_list == NULL) {
copy.dis_queue_list = lCreateList("disabled", lGetListDescr(t));
}
lAppendList(copy.dis_queue_list, t);
lFreeList (&t);
}
}
if (what_queue3) {
lFreeWhat(&what_queue3);
}
}
if (sconf_is_job_category_filtering()) {
copy.job_list = sge_category_job_copy(copy.queue_list, &orders, evc->monitor_next_run);
} else {
copy.job_list = lCopyList("", master_job_list);
}
/* no need to copy these lists, they are read only used */
copy.centry_list = master_centry_list;
copy.ckpt_list = master_ckpt_list;
copy.hgrp_list = master_hgrp_list;
/* these lists need to be copied because they are modified during scheduling run */
copy.share_tree = lCopyList("", master_sharetree_list);
copy.pe_list = lCopyList("", master_pe_list);
copy.user_list = lCopyList("", master_user_list);
copy.project_list = lCopyList("", master_project_list);
copy.rqs_list = lCopyList("", master_rqs_list);
copy.ar_list = lCopyList("", master_ar_list);
/* report number of reduced and raw (in brackets) lists */
DPRINTF(("Q:%d, AQ:%d J:%d(%d), H:%d(%d), C:%d, A:%d, D:%d, P:%d, CKPT:%d,"
" US:%d, PR:%d, RQS:%d, AR:%d, S:nd:%d/lf:%d \n",
lGetNumberOfElem(copy.queue_list),
lGetNumberOfElem(copy.all_queue_list),
lGetNumberOfElem(copy.job_list),
lGetNumberOfElem(master_job_list),
lGetNumberOfElem(copy.host_list),
lGetNumberOfElem(master_exechost_list),
lGetNumberOfElem(copy.centry_list),
lGetNumberOfElem(copy.acl_list),
lGetNumberOfElem(copy.dept_list),
lGetNumberOfElem(copy.pe_list),
lGetNumberOfElem(copy.ckpt_list),
lGetNumberOfElem(copy.user_list),
lGetNumberOfElem(copy.project_list),
lGetNumberOfElem(copy.rqs_list),
lGetNumberOfElem(copy.ar_list),
lGetNumberOfNodes(NULL, copy.share_tree, STN_children),
lGetNumberOfLeafs(NULL, copy.share_tree, STN_children)
));
if (getenv("SGE_ND")) {
printf("Q:%d, AQ:%d J:%d(%d), H:%d(%d), C:%d, A:%d, D:%d, "
"P:%d, CKPT:%d, US:%d, PR:%d, RQS:%d, AR:%d, S:nd:%d/lf:%d \n",
lGetNumberOfElem(copy.queue_list),
lGetNumberOfElem(copy.all_queue_list),
lGetNumberOfElem(copy.job_list),
lGetNumberOfElem(master_job_list),
lGetNumberOfElem(copy.host_list),
lGetNumberOfElem(master_exechost_list),
lGetNumberOfElem(copy.centry_list),
lGetNumberOfElem(copy.acl_list),
lGetNumberOfElem(copy.dept_list),
lGetNumberOfElem(copy.pe_list),
lGetNumberOfElem(copy.ckpt_list),
lGetNumberOfElem(copy.user_list),
lGetNumberOfElem(copy.project_list),
lGetNumberOfElem(copy.rqs_list),
lGetNumberOfElem(copy.ar_list),
lGetNumberOfNodes(NULL, copy.share_tree, STN_children),
lGetNumberOfLeafs(NULL, copy.share_tree, STN_children)
);
} else {
schedd_log("-------------START-SCHEDULER-RUN-------------", NULL, evc->monitor_next_run);
}
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_copy = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
scheduler_method(evc, &answer_list, ©, &orders);
answer_list_output(&answer_list);
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_run = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM7);
/* .. which gets deleted after using */
lFreeList(&(copy.host_list));
lFreeList(&(copy.queue_list));
lFreeList(&(copy.dis_queue_list));
lFreeList(&(copy.all_queue_list));
lFreeList(&(copy.job_list));
lFreeList(&(copy.acl_list));
lFreeList(&(copy.dept_list));
lFreeList(&(copy.pe_list));
lFreeList(&(copy.share_tree));
lFreeList(&(copy.user_list));
lFreeList(&(copy.project_list));
lFreeList(&(copy.rqs_list));
lFreeList(&(copy.ar_list));
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM7);
prof_free = prof_get_measurement_wallclock(SGE_PROF_CUSTOM7, true, NULL);
/*
* need to sync with event master thread
* if schedd configuration changed then settings in evm can be adjusted
*/
if (sconf_is_new_config()) {
/* set scheduler interval / event delivery interval */
u_long32 interval = sconf_get_schedule_interval();
if (evc->ec_get_edtime(evc) != interval) {
evc->ec_set_edtime(evc, interval);
}
/* set job / ja_task event flushing */
set_job_flushing(evc);
/* no need to ec_commit here - we do it when resetting the busy state */
/* now we handled the new schedd config - no need to do it twice */
sconf_reset_new_config();
}
/* block till master handled all GDI orders */
sge_schedd_block_until_orders_processed(evc->get_gdi_ctx(evc), NULL);
schedd_order_destroy();
/*
* Stop profiling for "schedd run total" and the subcategories
*/
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM6);
prof_total = prof_get_measurement_wallclock(SGE_PROF_CUSTOM6, true, NULL);
if (prof_is_active(SGE_PROF_CUSTOM6)) {
PROFILING((SGE_EVENT, "PROF: schedd run took: %.3f s (init: %.3f s, copy: %.3f s, "
"run:%.3f, free: %.3f s, jobs: %d, categories: %d/%d)",
prof_total, prof_init, prof_copy, prof_run, prof_free,
lGetNumberOfElem(*object_type_get_master_list(SGE_TYPE_JOB)), sge_category_count(),
sge_cs_category_count() ));
}
if (getenv("SGE_ND") != NULL) {
printf("--------------STOP-SCHEDULER-RUN-------------\n");
} else {
schedd_log("--------------STOP-SCHEDULER-RUN-------------", NULL, evc->monitor_next_run);
}
thread_output_profiling("scheduler thread profiling summary:\n", &next_prof_output);
sge_monitor_output(&monitor);
}
/* reset the busy state */
evc->ec_set_busy(evc, 0);
evc->ec_commit(evc, NULL);
/* stop logging into schedd_runlog (enabled via -tsm) */
evc->monitor_next_run = false;
/*
* pthread cancelation point
*
* sge_scheduler_cleanup_thread() is the last function which should
* be called so it is pushed first
*/
pthread_cleanup_push(sge_scheduler_cleanup_thread, (void *) &ctx);
pthread_cleanup_push((void (*)(void *))sge_scheduler_cleanup_monitor,
(void *)&monitor);
pthread_cleanup_push((void (*)(void *))sge_scheduler_cleanup_event_client,
(void *)evc);
cl_thread_func_testcancel(thread_config);
pthread_cleanup_pop(execute);
pthread_cleanup_pop(execute);
pthread_cleanup_pop(execute);
DPRINTF(("passed cancelation point\n"));
}
/*-------------------------------------------------------------------------*/
int main(int argc, char **argv)
{
int ret;
int my_pid;
int ret_val;
int printed_points = 0;
int max_enroll_tries;
static char tmp_err_file_name[SGE_PATH_MAX];
time_t next_prof_output = 0;
int execd_exit_state = 0;
lList **master_job_list = NULL;
sge_gdi_ctx_class_t *ctx = NULL;
lList *alp = NULL;
DENTER_MAIN(TOP_LAYER, "execd");
#if defined(LINUX)
gen_procList ();
#endif
prof_mt_init();
set_thread_name(pthread_self(),"Execd Thread");
prof_set_level_name(SGE_PROF_CUSTOM1, "Execd Thread", NULL);
prof_set_level_name(SGE_PROF_CUSTOM2, "Execd Dispatch", NULL);
#ifdef __SGE_COMPILE_WITH_GETTEXT__
/* init language output for gettext() , it will use the right language */
sge_init_language_func((gettext_func_type) gettext,
(setlocale_func_type) setlocale,
(bindtextdomain_func_type) bindtextdomain,
(textdomain_func_type) textdomain);
sge_init_language(NULL,NULL);
#endif /* __SGE_COMPILE_WITH_GETTEXT__ */
/* This needs a better solution */
umask(022);
/* Initialize path for temporary logging until we chdir to spool */
my_pid = getpid();
sprintf(tmp_err_file_name,"%s."sge_U32CFormat"", TMP_ERR_FILE_EXECD, sge_u32c(my_pid));
log_state_set_log_file(tmp_err_file_name);
/* exit func for SGE_EXIT() */
sge_sig_handler_in_main_loop = 0;
sge_setup_sig_handlers(EXECD);
if (sge_setup2(&ctx, EXECD, MAIN_THREAD, &alp, false) != AE_OK) {
answer_list_output(&alp);
SGE_EXIT((void**)&ctx, 1);
}
ctx->set_exit_func(ctx, execd_exit_func);
#if defined(SOLARIS)
/* Init shared SMF libs if necessary */
if (sge_smf_used() == 1 && sge_smf_init_libs() != 0) {
SGE_EXIT((void**)&ctx, 1);
}
#endif
/* prepare daemonize */
if (!getenv("SGE_ND")) {
sge_daemonize_prepare(ctx);
}
if ((ret=sge_occupy_first_three())>=0) {
CRITICAL((SGE_EVENT, MSG_FILE_REDIRECTFD_I, ret));
SGE_EXIT((void**)&ctx, 1);
}
lInit(nmv);
/* unset XAUTHORITY if set */
if (getenv("XAUTHORITY") != NULL) {
sge_unsetenv("XAUTHORITY");
}
parse_cmdline_execd(argv);
/* exit if we can't get communication handle (bind port) */
max_enroll_tries = 30;
while (cl_com_get_handle(prognames[EXECD],1) == NULL) {
ctx->prepare_enroll(ctx);
max_enroll_tries--;
if (max_enroll_tries <= 0 || shut_me_down) {
/* exit after 30 seconds */
if (printed_points != 0) {
printf("\n");
}
CRITICAL((SGE_EVENT, MSG_COM_ERROR));
SGE_EXIT((void**)&ctx, 1);
}
if (cl_com_get_handle(prognames[EXECD],1) == NULL) {
/* sleep when prepare_enroll() failed */
sleep(1);
if (max_enroll_tries < 27) {
printf(".");
printed_points++;
fflush(stdout);
}
}
}
if (printed_points != 0) {
printf("\n");
}
/*
* now the commlib up and running. Set execd application status function
* ( commlib callback function for qping status information response
* messages (SIRM) )
*/
ret_val = cl_com_set_status_func(sge_execd_application_status);
if (ret_val != CL_RETVAL_OK) {
ERROR((SGE_EVENT, cl_get_error_text(ret_val)) );
}
/* test connection */
{
cl_com_SIRM_t* status = NULL;
ret_val = cl_commlib_get_endpoint_status(ctx->get_com_handle(ctx),
(char *)ctx->get_master(ctx, true),
(char*)prognames[QMASTER], 1, &status);
if (ret_val != CL_RETVAL_OK) {
ERROR((SGE_EVENT, cl_get_error_text(ret_val)));
ERROR((SGE_EVENT, MSG_CONF_NOCONFBG));
}
cl_com_free_sirm_message(&status);
}
/* finalize daeamonize */
if (!getenv("SGE_ND")) {
sge_daemonize_finalize(ctx);
}
/* daemonizes if qmaster is unreachable */
sge_setup_sge_execd(ctx, tmp_err_file_name);
/* are we using qidle or not */
sge_ls_qidle(mconf_get_use_qidle());
sge_ls_gnu_ls(1);
DPRINTF(("use_qidle: %d\n", mconf_get_use_qidle()));
/* test load sensor (internal or external) */
{
lList *report_list = sge_build_load_report(ctx->get_qualified_hostname(ctx), ctx->get_binary_path(ctx));
lFreeList(&report_list);
}
/* here we have to wait for qmaster registration */
while (sge_execd_register_at_qmaster(ctx, false) != 0) {
if (sge_get_com_error_flag(EXECD, SGE_COM_ACCESS_DENIED, true)) {
/* This is no error */
DPRINTF(("***** got SGE_COM_ACCESS_DENIED from qmaster *****\n"));
}
if (sge_get_com_error_flag(EXECD, SGE_COM_ENDPOINT_NOT_UNIQUE, false)) {
execd_exit_state = SGE_COM_ENDPOINT_NOT_UNIQUE;
break;
}
if (shut_me_down != 0) {
break;
}
sleep(30);
}
/*
* Terminate on SIGTERM or hard communication error
*/
if (execd_exit_state != 0 || shut_me_down != 0) {
sge_shutdown((void**)&ctx, execd_exit_state);
DRETURN(execd_exit_state);
}
/*
* We write pid file when we are connected to qmaster. Otherwise an old
* execd might overwrite our pidfile.
*/
sge_write_pid(EXECD_PID_FILE);
/*
* At this point we are sure we are the only sge_execd and we are connected
* to the current qmaster. First we have to report any reaped children
* that might exist.
*/
starting_up();
/*
* Log a warning message if execd hasn't been started by a superuser
*/
if (!sge_is_start_user_superuser()) {
WARNING((SGE_EVENT, MSG_SWITCH_USER_NOT_ROOT));
}
#ifdef COMPILE_DC
if (ptf_init()) {
CRITICAL((SGE_EVENT, MSG_EXECD_NOSTARTPTF));
SGE_EXIT((void**)&ctx, 1);
}
INFO((SGE_EVENT, MSG_EXECD_STARTPDCANDPTF));
#endif
master_job_list = object_type_get_master_list(SGE_TYPE_JOB);
*master_job_list = lCreateList("Master_Job_List", JB_Type);
job_list_read_from_disk(master_job_list, "Master_Job_List",
0, SPOOL_WITHIN_EXECD,
job_initialize_job);
/* clean up jobs hanging around (look in active_dir) */
clean_up_old_jobs(ctx, 1);
execd_trash_load_report();
sge_set_flush_lr_flag(true);
sge_sig_handler_in_main_loop = 1;
if (thread_prof_active_by_id(pthread_self())) {
prof_start(SGE_PROF_CUSTOM1, NULL);
prof_start(SGE_PROF_CUSTOM2, NULL);
prof_start(SGE_PROF_GDI_REQUEST, NULL);
} else {
prof_stop(SGE_PROF_CUSTOM1, NULL);
prof_stop(SGE_PROF_CUSTOM2, NULL);
prof_stop(SGE_PROF_GDI_REQUEST, NULL);
}
PROF_START_MEASUREMENT(SGE_PROF_CUSTOM1);
/* Start dispatching */
execd_exit_state = sge_execd_process_messages(ctx);
/*
* This code is only reached when dispatcher terminates and execd goes down.
*/
/* log if we received SIGPIPE signal */
if (sge_sig_handler_sigpipe_received) {
sge_sig_handler_sigpipe_received = 0;
INFO((SGE_EVENT, "SIGPIPE received\n"));
}
#if defined(LINUX)
free_procList();
#endif
lFreeList(master_job_list);
PROF_STOP_MEASUREMENT(SGE_PROF_CUSTOM1);
if (prof_is_active(SGE_PROF_ALL)) {
time_t now = (time_t)sge_get_gmt();
if (now > next_prof_output) {
prof_output_info(SGE_PROF_ALL, false, "profiling summary:\n");
prof_reset(SGE_PROF_ALL,NULL);
next_prof_output = now + 60;
}
}
sge_prof_cleanup();
sge_shutdown((void**)&ctx, execd_exit_state);
DRETURN(execd_exit_state);
}
